Confound correction output contains voxels not present in the BOLD image

[mclooned]

Hi all,

When I run confound correction, I have noticed that the cleaned_timeseries nifti output file contains data not present in the bold image. I have tried playing around with different flags during preprocessing and using --nativespace_analysis during confound correction, but this has not had any impact.

As an example, below are sample images of a subject, with confound correction in native space.
Image 1 is the combined_bold_ref in native space.
Image 2 EPI brain mask that is used in confound_correction (from /main_wf/bold_main_wf/bold_native_trans_wf/....Brain_mask_EPI/).
Image3 combined_cleaned confound correction output.
Image 4 mask (2) overlaid on the bold image (1), showing the mask is larger than the bold brain.

How is the mask used in confound correction derived?
Is there any flag that I can used to avoid the confound output containing voxels not present in the bold image?

Thanks,
Daniel

[Gab-D-G]

Hello @mclooned , the brain is derived from the registration steps during preprocessing. The EPI is aligned to the anatomical image of the same session, and all sessions are aligned to a reference template which contains a brain mask. That brain mask is then propagated to each image.
So, if there's some part of the mask outside the brain as in your case, it is because there is not a perfect alignment between your image and the template. You should be able to infer where this originates from by inspecting the preprocessing QC report.
I can see in your case there is a mismatch where there are EPI distortions at the bottom of the brain, which is relatively common. If you don't apply a non-linear registration for the EPI, this won't be corrected.
If you want more detailed help to try fixing this, I need to see the related QC report for that image, and your preprocessing log report.

[mclooned]

Thanks Gabriel,

Alignment looks good to me. SyN reg was used for EPI to anatomical. See the QC outputs and log file below.

Is it the case that the mask used in confound correction is larger than the EPI simply because of the susceptibility distortion correction (SyN reg for EPI to anatomical)?

By replacing the brain mask used for confound (in this instance: /main_wf/bold_main_wf/bold_native_trans_wf/....Brain_mask_EPI/) with a mask created manually in 3DAutomask based on the combined_bold_ref (native), running confound results in an output which is more reflective of the bold image. However, maybe I am not using the correct bold image here - Is the susceptibility corrected bold output available in the preprocessing output workflow?

rabies_preprocess.log

[Gab-D-G]

I agree the alignment is definitely good enough, and it is not worth it to try improving further. I do believe though that despite the succesful registration, it was not sufficient to entirely correct for distortions using SyN. You can see in the EPI raw image that the location where there is a mismatch with the mask is where distortions are the most pronounced. I believe we can see some of that mismatch in the QC:

Now, unless your analyses are heavily dependent on that specific brain region (in which case, it may be best to review the MR acquisition protocol to reduce susceptibility distortions, since data-driven techniques have their limitations), you should be able to carry on analyses and bear in mind that results in those specific regions should be interpreted with caution. This is in general the case for whole-brain resting state analyses, in human fMRI as well.

Using another mask will not impact your results at this stage. The voxels which are outside the brain will primarily correspond to background random noise, and should not drive the results. If you derive a new mask for each image, you run the risk of having an inconsistent brain parcellation between images, because now you don't necessarily have the same number of voxels included for each image, and you may not be able to do commonspace analyses.

You may still do it however, if you like. You can derive your own mask and apply it onto the cleaned timeseries (confound correction ouputs), which will mask out these extra voxels. Or if you prefer images prior to this stage, the images in bold_datasink/commonspace_bold/ and bold_datasink/native_bold/ are both corrected for distortions. Those are the images you should use, and you won't need to dig into the main_wf/ folder.

Let me know if you have further questions!

[mclooned]

That makes sense. Thanks for your help, Gabriel!